Mid speed scada with DFE equalizer
By Miz.Wong, 2018
Miz.Wang@Hytera.com
Hytera Co,. Ltd.
System Info
--------------------------------------------------------------------------------------
Symbol Rate 10ksym/s
Band width 12.5kHz
Modulation type 16-QAM
Real information rate 35kBit/s
Baseband oversampling rate 8
IF oversampling rate 36
RF frequency( Only used for calculating VCO error and fd ) 350MHz
TX VCO frequency error 1ppm(3.5Hz)
TX NCO initial phase 0°
Channel info -- SNR 0dB
-- Doppler 2Hz
-- Path gains [0dB -22.3dB]
-- Path delays [0s 5e-6s]
RX NCO initial phase 60°
DFE taps 1 Forward 1 Backward
DFE forgetting factory 0.9
Frame structure |Training sequence|Information|
-------------------------------------------------------------------------------------
1.Clean up workspace
clear all;
close all;
clc;
2.Set system parameters
M = 16;
Rs = 10e3;
Rb = Rs * log2(M);
BB_OSR = 8;
FsBB = Rs * BB_OSR;
IF_OSR = 36;
FsIF = FsBB * IF_OSR;
Coefs = load('FilterCoefs.mat');
coefRRC = Coefs.coefRRC;
coefIF = Coefs.coefIF;
hMod = comm.RectangularQAMModulator( M, 'BitInput', true );
hDem = comm.RectangularQAMDemodulator( M, 'BitOutput', true );
T = 0.04;
nTSHead = 50;
tBB = 0 : 1/FsBB : T - 1/FsBB;
tIF = 0 : 1/FsIF : T - 1/FsIF;
txcVCO = 350e6;
txcPPM = 0.01;
txcFreq = 100e3;
txcFreq = txcFreq + (txcVCO / 1e6)*txcPPM;
txcPhi0 = 0;
rxcFreq = 100e3;
rxcPhi0 = pi/3;
fd = 2;
pdb = [0 -22.3];
tau = [0 5e-6];
RlChan = rayleighchan( (1 / FsIF), fd, tau, pdb);
hEVM = comm.EVM( 'ReferenceSignalSource', 'Estimated from reference constellation', ...
'ReferenceConstellation', constellation(hMod).' );
hMyDFE = fnDFECreate(1, 1, 0.9, hMod, hDem);
3.Transmitter
txBit = randi( [0,1], T*Rb, 1 );
txSym = step(hMod, txBit );
txSymUp = zeros( length(txSym)*BB_OSR, 1 );
txSymUp(1 : BB_OSR : end) = txSym;
txSymUp = txSymUp * BB_OSR;
txSymUp = [zeros(200,1); txSymUp; zeros(200,1);];
txSymUp = FilterConv( txSymUp.', coefRRC' ).' ;
txSymIF = zeros( length(txSymUp)*IF_OSR, 1 );
txSymIF(1 : IF_OSR : end) = txSymUp;
txSymIF = txSymIF * IF_OSR;
txSymIF = FilterConv( txSymIF.', coefIF' ).' ;
tIF = 0 : 1/FsIF : length(txSymUp)/FsBB - 1/FsIF;
txSymIF = txSymIF .* exp( 1j*( 2*pi*txcFreq*tIF' + txcPhi0 ) );
figure; plot(txSym, 'o-'); title('TX symbols');
hs = spectrum.periodogram; figure;psd(hs, txSymIF, 'Fs',FsIF, 'CenterDC',true);
4.Channel
rxSymIF = txSymIF;
rxSymIF = filter( RlChan, rxSymIF );
rxSymIF = awgn( rxSymIF, 0, 'measured' );
hs = spectrum.periodogram; figure;psd(hs, rxSymIF, 'Fs',FsIF, 'CenterDC',true);
5.Receiver
5.1 DDC & RRC
rxSymIF = rxSymIF ./ exp( 1j*( 2*pi*rxcFreq*tIF' + rxcPhi0 ) );
rxSymIF = FilterConv( rxSymIF.', coefIF' ).' ;
rxSymUp = zeros( length(rxSymIF) / IF_OSR, 1 );
rxSymUp = rxSymIF( 1 : IF_OSR : end );
rxSymUp = FilterConv( rxSymUp.', coefRRC' ).';
figure; plot(rxSymUp, '.'); title('Pre-VSR');
5.2 Symbol synchronization & Downsampling
for ii = 1:1:BB_OSR
avePwr(ii) = mean(abs(rxSymUp(ii:BB_OSR:end)));
end
[max T0] = max(avePwr);
rxSymUp = rxSymUp(201:end-200);
rxSym = rxSymUp(T0:BB_OSR:end);
fprintf( 'T0 = %d\r\n', T0 );
figure; plot(rxSym,'.'); title('Pre-DFE');
5.3 Decision Feedback Equalization (DFE)
[hMyDFE, rxSymDFE, vyDcs] = fnDFE( hMyDFE, rxSym, [1:nTSHead], txSym );
figure; plot(rxSymDFE(nTSHead+1:end),'.'); title('Post-DFE');
6.DFE performance
BitNeq = step(hDem, rxSym);
BERDFE = (sum( vyDcs(nTSHead*log2(M)+1:end)' ~= txBit(nTSHead*log2(M)+1:end) ) / length(txBit(nTSHead*log2(M)+1:end))) * 100;
BERORG = (sum( BitNeq(nTSHead*log2(M)+1:end) ~= txBit(nTSHead*log2(M)+1:end) ) / length(txBit(nTSHead*log2(M)+1:end))) * 100;
fprintf( 'BER : \r\nWithout DFE : %f%%\r\nWith DFE : %f%%\r\n', BERORG, BERDFE );
EVMDFE = hEVM(rxSymDFE(nTSHead+1:end));
EVMORG = hEVM(rxSym(nTSHead+1:end));
fprintf( 'EVM : \r\nWithout DFE : %f%%\r\nWith DFE : %f%%\r\n', EVMORG, EVMDFE );
figure; stem(real(hMyDFE.e));hold on;stem(imag(hMyDFE.e));title('Equalizer error');
figure; plot(abs(txSym./rxSym),'x-');hold on;plot(abs(txSym./rxSymDFE),'o-');
title('Magnitude error without and with DFE');
figure; plot(phase(txSym./rxSym),'x-');hold on;plot(phase(txSym./rxSymDFE),'o-');
title('Phase error without and with DFE');
Performance of BER
This figure shows the BER performance in different SNR :
More about DFE
DFE is a decision feedback structure, shown below:
Currently, my code only support symbol-space DFE, so K is always 1.
'Weight setting' part can be LMS, RLS, Kalman and other algorithms. Currently, only RLS is supported.
The core formula of RLS is :
K is kalman gain matrix, P is inverse correlation matrix, u is the input vector, e is the error vector, lamda is fogetting factory, H denotes the Hermitian transpose, * denotes the complex conjugate.